Hi everyone, I'm Hugo, a French engineer of 25. I've been working on GreenKube for the last 10 months in my free-time.

I work in a 15-person IT services company and I didn't found a FinGreenOps tool for K8s that is free, easy-to-use and not overkill for small/medium clusters. So I started developping GreenKube to provide energy consumption and CO2 emissions of every pod in your cluster, with actionnable recommendations to reduce both emissions and cloud bills.

How it works: - Collects CPU, memory, network, and disk metrics from Prometheus - Maps each node's instance type to a power profile (min/max watts per vCPU from SPECpower benchmarks, same methodology as Cloud Carbon Footprint) - Linearly interpolates power draw based on actual CPU utilization - Multiplies by PUE and grid carbon intensity (real-time via Electricity Maps or a configurable default)

It comes with an 8-type recommendation engine (zombie pod detection, CPU/memory rightsizing, etc.), a built-in SvelteKit dashboard, a REST API (FastAPI), a CLI, and a pre-built Grafana dashboard. Everything is packaged into a single Helm chart and exports native Prometheus metrics using standard K8s labels (namespace, pod, node), making sustainability data easy to join to your application performance stats.

It should support Azure, OVH, AWS, GCP and Scaleway out of the box. On-prem clusters work too with manual zone labels. I've tested Azure, OVH and on-prem, but not AWS, GCP and Scaleway yet, so if you find a mistake, don't hesitate to report it!

Try it in 30 seconds with docker (no cluster needed): docker run --rm -p 9000:9000 greenkube/greenkube:0.2.11 demo --no-browser --port 9000

Or explore the live demo: https://demo.greenkube.cloud

To install full version, follow github readme or the doc at https://www.greenkube.cloud/

Current limitations (and where I'd love feedback):

- The energy model is currently CPU-only. Memory, disk, network, and GPU are collected as metrics but not yet factored into the energy estimate. - CO2 estimation uses a TDP-based linear interpolation model (same approach as Cloud Carbon Footprint). I'd really appreciate suggestions on improving accuracy and include RAM, disk usage, etc... in the estimation. - No GPU support yet.

The project is Apache 2.0 licensed. Built with Python (async), FastAPI, SvelteKit, and Helm. I'd be grateful for any feedback, on the tool itself, the estimation methodology, feature priorities, or anything else.

Submissions

Show HN: GreenKube – Open-source K8s cost and CO2 optimization engine

Ask HN: What's your favorite K8s tool in 2026?

comments